README

DataSet

适用于机器学习算法训练的多功能dataSet。

创建DataSet

为了创建用于Zeeml机器学习的DataSet，您需要指定一个源：要么是一个CSV文件，要么是一个数组

从CSV文件创建DataSet

$dataSet =  DataSetFactory::create('/path/to/csv', ['name', 'Gender'], ['Height]);

在标题中设置的键（CSV文件的第一行）用作DataSet的键

从数组创建DataSet

$dataSet =  DataSetFactory::create(
    [
        ['name' => 'Zac',    'gender' => 'Male',    'height' => 180],
        ['name' => 'Emily',  'gender' => 'Female',  'height' => 177],
        ['name' => 'Edward', 'gender' => 'Male',    'height' => 175],
        ['name' => 'Mark',   'gender' => 'Male',    'height' => 183],
        ['name' => 'Lesly',  'gender' => 'Female',  'height' => 170],
    ]
);

其他任何数组格式都会抛出异常

指定输入和输出

在调用其他任何方法之前必须先调用prepare方法，否则将抛出异常。

$mapper = new Mapper(['name', 'gendre'], ['height']);
$dataSet->prepare($mapper);

其中 ['name', 'gendre'] 是用作输入的索引，['height'] 是用作输出的索引。

可以从条目中选择任意数量的输入和输出

如果键不存在，将抛出异常。

操作DataSet

为了操作和更改DataSet的值（清洗、重命名等），您可以应用一个"策略"。

在创建Mapper时调用策略。每个列可以定义多个策略

$dataSet = DataSetFactory::create(
      [
          [180, 'Male'],
          [177, 'Female'],
          [170, ''],
          [183, 'Male'],
      ]
);
$mapper = new Mapper(
    [
        0 => [Policy::replaceWithAvg(), Policy::rename('height')], 
    ], 
    [
        1 => [Policy::skip()]
    ]
);
$dataSet->prepare($mapper);

###支持策略

Policy::skip() : 如果对应索引的值是空的（NULL、false、''），则整行将被跳过

示例

$data = [
    [1, 2, 3],
    [4, null, 5],
    [6, 7, null],
    [null, 8, 9],
];

$dataSet =  DataSetFactory::create($data);
$mapper = new Mapper([0, 1 => Policy::skip()], [2 => Policy::skip()]);
$dataSet->prepare($mapper);

will use the following Inputs/Outputs :

Inputs:                                             
[                                                 
    [1, 2],                                        
    [null, 8], //No policy applied on 0           
]                                                

Outputs:    
[
    [3],
    [9],
]

Policy::replaceWith() : 如果对应索引的值是空的（NULL、false、''），则将其替换为给定的值

示例

$data = [
    [1, 2, 3],
    [4, null, 5],
    [6, 7, null],
    [null, 8, 9],
];

$dataSet =  DataSetFactory::create($data);
$mapper = new Mapper([0, 1 => Policy::replaceWith('Unknown')], [2 => Policy::replaceWith(-1)]);
$dataSet->prepare($mapper);

will use the following Inputs/Outputs :

Inputs:                                           
[                                                  
    [1, 2],                                          
    [4, 'Unknown'],                                  
    [6, 7],                                          
    [null, 8], //No policy applied on 0              
]                                                  

Outputs:                                          
[
    [3],
    [5], 
    [-1],
    [9]
]

Policy::replaceWithAvg() : 将空值替换为从原始DataSet计算出的该列的平均值

示例

$data = [
    [1, 2, 3],
    [4, null, 5],
    [6, 7, null],
    [null, 8, 9],
];

$dataSet =  DataSetFactory::create($data);
$mapper = new Mapper([0 => Policy::replaceWithAvg(), 1 => Policy::skip()], [2 => Policy::replaceWithAvg()]);
$dataSet->prepare($mapper);

will use the following Inputs/Outputs :

Inputs:                                                              
[                                                                    
    [1, 2],                                                            
    [6, 7],                                                                                    
    [2.75, 8], // Avg(0) = 1 + 4 + 6 + 0 = 11 / 4 = 2.75               
]                                                                    

Outputs:
[
    [3],
    [-1],
    [9],
]   
                                                                ]

Policy::replaceWithMostCommon() : 将空值替换为最常见值（出现次数最多的值）。如果多个值具有相同的频率，则随机选择一个。

示例

$data = [
    [1, 2, 3],
    [1, null, 5],
    [6, 7, null],
    [null, 8, 9],
];

$dataSet =  DataSetFactory::create($data);
$mapper = new Mapper([0=> Policy::replaceWithMostCommon(), 1 => Policy::skip()], [2]);
$dataSet->prepare($mapper);

will use the following Inputs/Outputs :

Inputs:                                 
[                                        
    [1, 2],                               
    [6, 7],                               
    [1, 8],                                
]                                      

Outputs:
[
    [3],
    [null],
    [9],
]

Policy::custom() : 创建自己的策略

可调用函数仅在值为空时调用。可调用必须

接受一个引用参数作为每次迭代中列的值
接受一个参数作为行
返回true以保留行，返回false以跳过行

示例

$data = [
    [180, 'Male'],
    [177, 'Female'],
    [170, ''],
    [183, 'Male'],
];

$dataSet =  DataSetFactory::create($data);

$genderCleaner = function(&$value, $line) {
    if ($line[0] > 175) {
        $value = 'Male' ;
    } else {
        $value = 'Female';
    }
    
    return true;
}

$mapper = new Mapper([0], [1 => Policy::custom($genderCleaner)]);
$dataSet->prepare($mapper);

will use the following Inputs/Outputs :

Inputs:                                 
[                                        
    [180],                               
    [177],                               
    [170],                                
    [183],                                
]                                      

Outputs:
[
    ['Male'],                               
    ['Female'],                               
    ['Female'],                                
    ['Male'],
]

重命名DataSet键

您可以重命名DataSet键

$data = [
    ['Zac',    'Male',    180],
    ['Emily',  'Female',  177],
    ['Edward', 'Male',    175],
    ['Mark',   'Male',    183],
    ['Lesly',  'Female',  170],
];    
    
$dataSet =  DataSetFactory::create($data);
    
$mapper = new Mapper([0, 1], [2]);
$dataSet->prepare($mapper);
    
$dataSet->rename([0 => 'Name', 1 => 'Gender', 2 => 'Height']);

and the inputs/outputs matrices used are :

Inputs :
[
    ['Name' => 'Zac',    'Gender' => 'Male'],
    ['Name' => 'Emily',  'Gender' => 'Female'],
    ['Name' => 'Edward', 'Gender' => 'Male'],
    ['Name' => 'Mark',   'Gender' => 'Male'],
    ['Name' => 'Lesly',  'Gender' => 'Female'],
]

Outputs :
[
    ['Height' => 180],
    ['Height' => 177],
    ['Height' => 175],
    ['Height' => 183],
    ['Height' => 170],
]

zeeml / dataset

维护者

详细信息